OpenAI moved on two fronts within the past day. GPT-5.6 Luna is now the default model for free ChatGPT users, and its API price dropped roughly 80 percent — from $1 to $0.20 per million tokens — a direct response to lower-cost Chinese rivals like DeepSeek and Moonshot AI winning users on price. At the same time, reports say OpenAI’s S-1 — the formal paperwork a company files before selling shares to the public — is imminent, setting up one of the most consequential tech IPOs ever. The company also previewed an “Ultrafast” mode that runs GPT-5.6 Sol up to 14 times faster, powered by Cerebras chips.
Anthropic, the company behind Claude, has reportedly turned profitable for the first time — a milestone in an industry where losses are the norm. In the same breath, all seven cofounders publicly pledged 80 percent of their personal wealth to philanthropy, worth roughly $5.4 billion each at the company’s current $380 billion valuation. For a communications leader, this is a masterclass pairing: a hard business proof point wrapped in a values story, landing in the same news cycle as rivals cut prices to defend market share.
Elon Musk’s xAI (now folded into SpaceX and increasingly branded “SpaceXAI”) released Grok 4.6, which early testers rank among the strongest knowledge-work models available — and notably cheap to run. It powers Grok Bot, launched in beta on August 11: a team of always-on AI “teammates,” each with its own cloud computer, that you message like a colleague. A bot signs into your tools, works across your apps the way a person would, and returns with a finished result. Musk says Grok 4.7 is only weeks away, underscoring how fast the release cycle has become.
Jeff Dean and Sanjay Ghemawat — the engineering duo whose distributed-systems work let Google Search scale for two decades — are leaving after more than 25 years, and markets treated it as a crisis. Exponential View’s new analysis argues the opposite: the departures reveal where we are in the AI investment cycle. Google’s TPU chips now earn such attractive returns serving today’s production models that open-ended research struggles to justify the compute. The piece lays out seven charts and four signals that would show the infrastructure cycle turning.
Anthropic’s decision to embed invisible watermarks in all text generated by new Claude models — plus signed provenance metadata on images under the C2PA standard — continued to dominate discussion this weekend. The move answers Article 50 of the EU AI Act, which requires AI-generated content to be identifiable, but Anthropic is applying it worldwide, not just in Europe. Users are pushing back hard against the idea that their AI-assisted work becomes detectable. For anyone managing corporate reputation, this is the live case study on transparency mandates colliding with user expectations.
Google cuts Gemini 3.7 Flash pricing in halfGoogle dropped a sharper, cheaper workhorse model overnight as the industry-wide price war intensifies. The Neuron notes Google, OpenAI, and DeepSeek all shipped model updates the same day.
OpenAI introduces “Computer History” for ChatGPTChatGPT can now remember and reference what it has done on your computer across sessions, per AI Valley — a step toward persistent AI assistants that know your working context.
OpenAI appoints Dali Rajic as Chief Revenue OfficerA new revenue chief to lead the global sales organization — notable timing given the reported S-1, and days after COO departure reports surfaced in Lore’s newsletter.
Cursor is now reported to be part of SpaceXAI Valley reports the popular AI coding editor has been folded into Musk’s SpaceX/xAI orbit, with The Neuron previously flagging a possible Grok rebrand.
Amazon trains AI on Twitch streams — opt-out, not opt-inTwitch added a setting to opt out of AI training; an Amazon executive conceded that “if it was opt in, nobody would opt in.” A creator-relations flashpoint in the making.
The dedicated Substack tools were unavailable this run, so newsletters were pulled from your Inoreader newsletter folders instead. It was a quiet weekend window — these are the editions that arrived in the past 24 hours.
A practical roundup of AI tools for making eye-catching graphics: Adobe’s conversational design tools let you simply describe what you want, with access to dozens of Adobe apps behind the scenes.
Canva’s expanding AI feature set works well as a “headquarters” when juggling multiple design projects — Caplan says an updated Canva guide is coming.
Timely flag: Flux 3, the new video model from Black Forest Labs, is free to try only through August 16 — today. It generates clips up to 20 seconds with optional audio, from a text prompt or start/end frame images.
Jeff Dean and Sanjay Ghemawat’s departures were read as crisis; the authors argue they actually reveal Google’s shifting capital allocation — TPUs earn so much serving production models that open-ended research can’t clear the return hurdle.
Seven charts track how Google’s compute has moved away from research and what Google Cloud’s economics reveal about infrastructure demand.
The essay closes with four signals to watch for the AI infrastructure cycle finally turning — a useful framework amid this week’s capex debate.
Fred Schott, creator of the Astro web framework, is building Flue — described as “like Claude Code, but 100% headless and programmable” — where an AI agent is just a JavaScript function that re-renders before every model call.
He’s added React-style “hooks,” arguing nobody has yet built “the React for agents”; Sierra CEO Bret Taylor’s line that we’re in “the jQuery era of agents” frames the piece.
Plain-English takeaway: developers are racing to standardize how AI agents are built, the way web frameworks standardized websites — worth tracking as agents enter comms workflows.
Sommer’s media-and-politics dispatch on the administration’s posture toward the Supreme Court — useful context for the US political-media environment your US colleagues are navigating.
The latest instalment in Mounk’s series on Francis Fukuyama’s intellectual evolution, this time on state formation — long-read weekend material rather than news.
From Your AI Feeds
Inoreader AI Folder.
3 articles landed in your “AI” folder in the last 24 hours — a quiet Saturday window dominated by agent frameworks and the Google talent story. Friday’s heavier flow (Grok 4.6, watermarks, GPT-5.6 pricing) was covered in yesterday’s edition and resurfaces above where still developing.
The Automated’s weekend digest, recapping the week’s biggest AI stories in one skimmable email. Earlier editions this week flagged Anthropic’s research on what happens when AI agents are forced to work together on conflicting tasks (“messier than expected”) — a theme echoed across the agent-workflow stories in today’s briefing.
Azeem Azhar and Hannah Petrovic’s analysis of the Jeff Dean and Sanjay Ghemawat departures, summarized in full above.
See also · Top Stories · Substack Highlights
Discoveries
Workflows & Tool Watch.
Last day: Flux 3 video generation is free through August 16
Black Forest Labs’ Flux 3 generates video and synchronized audio in a single pass — up to 20-second clips from a text prompt, or from uploaded images used as starting and ending frames. Replicate calls its direction-following the best they’ve tested. If you want to evaluate it for Red Book podcast promos or social clips, today is the last free day.
Multiple newsletters this week reported Anthropic making auto-mode the default in Claude Code — meaning it proceeds through multi-step work without asking permission at every step. Future Tools and The Neuron both flagged the change. In plain terms: the tool now behaves less like a assistant awaiting sign-off and more like a delegated staffer. Worth understanding before your team leans on it for anything sensitive — and a good prompt to review what approvals you want to keep manual in your own Cowork sessions.
A newsletter-production workflow built on Claude Desktop and MCP servers
A detailed public write-up shows how one builder assembled an AI newsletter pipeline using Claude Desktop, a set of MCP servers (the connectors that let Claude reach outside apps), and Langfuse to track and refine outputs over time. It mirrors how this very briefing works — and offers ideas for adding quality-tracking to any recurring AI-generated deliverable your comms team produces.
RelevantClaude Desktop · MCP · Inoreader · Media monitoring
n8n community nodes for Obsidian connect your vault to automations
The n8n-nodes-obsidian project (by Matt Williams, “technovangelist”) lets n8n workflows read from and write to an Obsidian vault directly. Practical uses: auto-filing meeting summaries into your vault, or piping flagged Inoreader articles into a daily note. Pairs naturally with the community’s newer web-capture workflows.
The AI Maker’s explainer sorts AI automation into three levels — single prompts, chained multi-step workflows, and autonomous agents — with guidance on when each is enough. A clean mental model for deciding which team tasks deserve a full agent versus a well-crafted prompt, and a companion read to WhatPlugin.ai’s “Should you automate it?” issue on what to leave alone.
No major new Tencent AI headlines broke in the past 24 hours, but the global availability of Tencent Hy3 continues to draw coverage: free worldwide access via WorkBuddy until August 31, plus availability through Tencent Design Miora and Tencent Cloud TokenHub. The 90 percent internal task-completion figure for agent work is the stat reporters keep citing.
IG’s preview of Tencent’s Q2 2026 results centers on AI capital spending against a gaming slowdown. Expect journalist queries on AI monetization timelines, Hy3 adoption metrics, and how Tencent’s capex compares with the $863 billion figure Exponential View pegs for the seven largest AI infrastructure builders this year.
InfoWorld’s piece on an ex-OpenAI researcher launching a new model for Tencent is being read alongside the week’s Google-exodus coverage — the AI talent-flow narrative now includes movement toward Chinese platforms, a useful counterpoint if the “China lagging” framing comes up.
The globally launched Hunyuan 3D Engine — advanced 3D creation tools for creators — pairs with the Hy3 story as evidence of Tencent shipping practical creative AI, not just foundation models. Good proof points for creative-industry press.